Logistic regression with missing values in the covariates (Q1327701): Difference between revisions

Assume logistic regression of the binary variable \(Y\) on covariates \(X_ 1\) and \(X_ 2\) where missing values of \(X_ 2\) can occur. MAR and MDX state, that the probability of missing \(X_ 2\) depends, respectively, only on the values of \(Y\) and \(X_ 1\) and only on the value of \(X_ 1\). Maximum likelihood estimation is studied under MAR. Pseudo maximum likelihood estimation, involving replacement of a nuisance parameter by a consistent estimate, is studied under \(MDX\). Finally, under MAR, the filling method is considered. Here missing values are estimated prior to using maximum likelihood estimates. The main evaluation of procedures is asymptotic with only simulation studies for finite sample sizes. Finally, some more general models are discussed.